liquid xenon calorimeter analysis r.sawada on behalf of the meg lxe analysis group 17/feb/2009
TRANSCRIPT
Liquid Xenon Calorimeter AnalysisR.Sawada
on behalf of the MEG LXe analysis group17/Feb/2009
Introduction
3
Strategy of LXe analysis for MEG
• We develop independent two analysis lines to cross-check each other.
• Two different waveform(WF) digitizers. DRS : PSI custom chip, fast sampling, and TRG : commercial chip, slower sampling.
• Different reconstruction algorithms.
• Cross checking is done at lower level analyses (PMT calibration, linearity, resolution..), as well as higher level (physics) analysis.
• This strategy works well to catch problems quicker and understand better.
4
Strategy of LXe analysis for MEG
WF analysis - noise reduction - charge, time estimationPMT analysis
Reconstruction - energy - position - time
Performance studyPMT calibrationLiquid xenon property
P.D.F.(probability function)
Physics data reconstruction
Physics analysis
Feedback
This talkAngela’s calibration talk
5
Status of 2009 data analysis
• At this moment, we are trying the same methods of analysis done for 2008 as the first step.
• We are preparing for further improvements.
• Alpha particle data in gas phase xenon, which was used to estimate quantum efficiencies(Q.E.) of PMT in 2008 analysis, have not been taken. So, 2009 data analyses shown in this slide were done by using Q.E. measured in 2008.
6
Outline
• Waveform
• Reconstruction
• Energy
• Time
• Position
• Pileup
• Efficiency
• Summary
In each section, 2008 result and status of 2009 analysis are shown.
XEC waveform analysis for DRS
We used DRS2 in 2008, and upgraded to DRS4 in 2009
8
Noise reduction
DRS2
DRS4
Cell pedestal(offset voltage of sampling cells)
Noise templateCorrelated with trigger
Am
plitu
de (
mV
)
Appears after several days after hardware calibration
Cross talk with clock WF
Template of noise is made for each channel by averaging many random trigger events.The template is subtracted in analysis.
In DRS4, no step at the last 64 cells and no crosstalk of clock are seen.
9
Noise reduction
2008
Before reductionAfter reduction
Noise is reduced by offline analysis from ~0.7 mV to 0.4 mV RMS
In 2009 (DRS4), noise level is lower (0.34 mV)after the noise reduction
10
Contribution to energy
150 ns
We use normal integration (NI), and high pass filtered (HP) waveforms for charge integration.Physics analysis is done with HP.
NI
HP
Energy distribution in random trigger events.
NI
HP
Contribution of noise to energy is small enough.In 2008,
0.7 % before noise reduction0.4 % after noise reduction
In 2009, it is 0.25 % after noise reduction.Thanks to lower noise level + higher light yield.
11
Saturated PMT
Saturation voltage of DRS2 are chips dependent, and can vary in time.That of DRS4 are common in all the chips and stable.
Charge of saturated PMT is estimated by using a known relation between charge and ToT (time over threshold)
2009
2008*
* More accurately, baseline of splitter can change, and it affects saturation voltage.
ToT
12
Timing
Time of each PMT is estimated by Constant-Fraction method in offline analysis.
DRS2
DRS3
Time of different chips are synchronized by using distributed clock waveforms.
before synchronization
PMT waveform
Clock waveform
Energy
14
Energy reconstruction
✦Weighted photon sum
charge gain Q.E. correction for PMT coverage fraction (fixed)
✦Correction of - Non-uniform response in the detector - Variation of light yield
✦Scale to energy. (Single factor)
Several other methods, studies are on going- Linear fit (estimation of the best weighting factors)- Gamma position dependent weighting- Estimating energy by each PMT, then taking weighted mean
works well for the large prototype. precise knowledge of optical property needed
15
Non-uniformity and correction
2008- 21/Nov
200821/Nov -
2009
0.14% σCW
(uncorrected)
CW (corrected)Uniformity
Relative response map in LXe entrance window, before correction (color code represents response.)
Non-uniformity is measured from CW data.Response depends on purity and Q.E.
estimation.
DRS
This non-uniformity also exists in MC.Response is larger in areas close to top or bottom faces.
16
Light yield variation and correction
Purification re-started
Energy scale measurement
Absolute scale of energy was measured by 55 MeV gamma in Pion runs.Variation was monitored by Pion, Cosmic ray, AmBe and CW data, and corrected.
2008
Uncertainty of energy scale : 0.4 %
Main uncertainties from gain variation in pion runs, and peak estimation of cosmic ray data.
(In 2009, those uncertainties are not present.)
17
Light yield history, 2008 and 2009
DRS
Physics data taking
2008
2009
± 1 %
40 %
Light yield in 2009 is very stable.No correction is needed.
Light yield is same as the large prototype.(i.e. maximum in our experience)
Scale of this figure is factor 2.2 larger than left top figure.(Light yield was increased)
2008 and 2009 together
DRS
18
Energy resolution
Resolution map in entrance window
σup=2.0% for deep(>2cm), 3.0% (1~2cm), 4.2% (0~1cm)
π0 55MeV
Energy resolution is depth dependent.
Fiducial volume
Average resolution (sigma of higher energy side)
• Resolution is measured from 55 MeV gamma data• Difference of pedestal distribution in pion and muon data (much more pileup gamma in pion data) is taken into account.
Pileup, much more in pion data
Interaction before xenon
2008
19
Energy resolution in 2009
• 1.95% σ @ 52.8MeV
CEX
CW
Noise contribution
Energy resolution as a function of energy.
Signal
20
Checking energy analysis by muon data
• Fitting gamma spectrum in muon runs by MC energy deposit.
• Free parameters ( #RD+AIF, #pileup, energy resolution, energy scale )
• Fit result (RD + AIF + Pileup)
Energy resolution and scale can be checked by using
muon data
21
Checking by using muon data
Pion and muon data are consistent.
Pileup
Single gamma
signalResolutions at various positions
Scale in 2008 is 1 (i.e. consistent with pion data)2009(200u degrader), and 2009(300u degrader) are little higher because of a mistake of configuration parameter, to be reprocessed.
*
* Scale factor for NI was used, while analysis is done by HP (see page 10)
Time
23
Time reconstruction
• Time reconstruction is done by fitting or averaging PMT time• Taking into account time-walk, delay due to photon statistics, cable length, path length from gamma conversion point to each PMT and effect of indirect photons.• Calibrations constants (PMT time offsets, time-walk correction factors...) are obtained from pion runs.
24
Time resolution measurement
γ γ
D : depth resolution, c’ : effective speed of light
measurements
× 2
“Intrinsic” resolution
“Practical” resolution
25
Resolution
• “Intrinsic” resolution
• 45 psec at 55 MeV
•“Practical” resolution
• 78ps @55MeV, 61ps @83MeV in summer 2008
• 68ps @55MeV in winter 2008, due to higher light yield
Intrinsic“Practical” resolution as a function of energy
In 2009, “intrinsic” resolution is same as 2008, but “practical” is worse (169 psec); to be understood.
Possibly it is due to a jitter of electronics. The jitter is chip dependent, and it was measured to be much larger(up to 155 ps) between LXe and pre-shower chips, than that between LXe and TC chips (up to 102 ps). Note that the effect is smaller for physics analysis. If we subtract the jitter, LXe resolution is sqrt(169^2 - “<155”^2) = “>67” psec, which is almost same as winter 2008. See backup slide for details.
2008
2008
26
Drift of relative time to positron
Stability after correction < 20 ps
Purificatoin
Begin 2008End 2008
RD peak in low intensity muon runs
Drift of time during 2008 physics runs
300 psec
2009
No drift is seen in 2009
Waveform was getting wider in 2008
Constant-Fraction time depend on pulse
shape
➔ Variation was well monitored and corrected
2008
Position
28
Position reconstruction and resolution
Data MC
Slits
✦ Reconstruction : Fitting PMT output distribution on the inner face, by a sold angle function.
✦ Resolution measurement
− π0 runs with 1cm lead slits in front of the detector
− Compare the width of shadow of slits with MC
− Result
− The width is 1.8 mm wider than MC (resolution is known) in quadrature
σxy~5 mm, σdepth~6 mm
Double GaussianCore ~75%, σ4mmTail ~25%, σ9mm
MC
2008
29
Triggered area
Lead slits
Position resolution in 2009
Sigma : 6~7.5 (last year 6.7~7.2mm)
A similar measurement was done in 2009, and resolution is same as 2008.
gas alpha data is not taken yet for 2009, so Q.E. in 2008 is used in the analysis.
Pileup analysis
31
Analysis for pileup events
Finding pileup from light distribution pattern
→ Estimating energy without using PMTs around the pileup (resolution could be worse than usual way)
→ Replacing PMT output only around the pileup with expectation from main gamma
→ Usual reconstruction
We don’t throw away pileup events.
Before replacementPMT output (development view of the detector)
32
Finding pileup from light distribution pattern
→ Estimating energy without using PMTs around the pileup (resolution could be worse than usual way)
→ Replacing PMT output only around the pileup with expectation from main gamma
→ Usual reconstruction
We don’t throw away pileup events.
After replacementPMT output (development view of the detector)
Analysis for pileup events
33
Muon data with pileup elimination
Pileup identification is done by two methods: time and space distribution of PMT outputs
2008 muon data, gamma energy spectrum
Black : acceptance, cosmic ray rejectedRed : rejected pileup events identified only by time methodBlue : After pileup contribution elimination
Inefficiency to the signal : 5.5 ± 2.5 %
signal
BG gamma rate is significantly decreased around signal energy.
34
Efficiency
✦ Detection efficiency
✦ Three methods to measure efficiency
− signal MC
− π0 2 gamma rate, at NaI single trigger
− muon runs, gamma rate
✦ Position dependence, and event distribution in muon data are taken into account
✦ εdet = 66 % ([46, 60] MeV)
✦ Analysis efficiency (Pileup, and cosmic ray cut)
εγ = (63 ± 4) %
consistent in 5 % (rel.)
γγ tag
?
Interaction before xenon = inefficiency
NaI
Energy spectrum of NaI trigger data
46 MeV
35
Performance
2008 2009 (Tentative)
Gamma Energy (%)Gamma Timing (psec)Gamma Position (mm)Gamma Efficiency (%)
2.0/3.0/4.280
5-6 (~7.5 mrad)63
2.0/NA/NA> 67
5-6 (~7.5 mrad)NA
Tentative performance in 2009 is already similar to 2008.
*
* For different depth, 2>cm / 1-2 cm / 0-1 cm
**
**
** To be estimated
36
Future improvements
• Purity of xenon(i.e. optical property) was stable in 2009
• Optimization of analysis for a single set of optical parameters must work.
• Even more detailed study on optical properties. ➔ The better Q.E. estimation ➔ The better position, time and energy estimation.
• Improvements of energy reconstruction
• Better equalization of PMT output to obtain uniform response before a posterior correction.
• Better weight for summing PMT output (LinearFit method worked very well for the large prototype)
• New algorithms to decrease fluctuation from low output PMTs. (energy estimation by each PMT)
QuickTime™ and aMotion JPEG OpenDML decompressor
are needed to see this picture.
LXe convection seen in a glass cryostat at KEK
• Ideas to improve timing performance of DRS (➔ Stefan’s talk)
• More stable operation of the detector by more powerful cooling. The large prototype was operated 100% by a refrigerator during the measurement, while we use also LN2 for the final detector. Effect of convection to resolutions is not known: Convection may look like position and time dependent scattering.
37
Summary
• Good performance was achieved for 2008 data analysis.
• Noise level of WF is low enough.
• Time and position resolutions are good.
• A few % non-uniformity of energy, and 40% change of light yield are well monitored and corrected. 2 % energy resolution was achieved.
• BG gamma rate was decreased by eliminating pileup contribution, instead of just throwing away events.
• 2009 performance is estimated by using the same method with 2008. Performance is almost same as 2008. Further improvements will be tried.
• Hardware modifications (Cooling, DRS4) are planed.
• Several ideas to improve analysis, to be developed and verified.
Backup
Effect of electronics for 2009 time resolution measurement
40
Measurement electronics contribution
• Square pulse into inputs of splitters, going to “Xenon inner”, “Pre-Shower”.
• One pair (two chips) is used at each run.
• Similar measurement was done between Xenon and TC (in Backup slides)
Pulse generator
xenon inner
Pre-shower
TC NIM
41
Result
165 psec
One entry is a pair of a XEC and pre-shower chip
On the same chip : 30 psecOn same pre-shower board : 104
Leading edge time
XEC Inner chips
83 psec
165 psec
< 155 psec
(This jitter is specific to the pi0 data. Not same as physics data)
42
Time resolution measurement (55 MeV)
γ γ
D : depth resolution, c’ : effective speed of light
measurements
43 psec
x 2
202 psec <155 psec60 psec93 psec
>68 psec
43 psec
~60 psec (MC)
>53 psec
is same as 2008
Speed of scintillation light in LXe
44
Measurement by using alpha data
• Vary alpha source position to obtain different distances.
• Use only PMTs around the end of the wire to minimize the systematic effect due to reflection and scattering.
• Wide distance range [0, 48.7]cm
V = (Di - di ) / (Ti - ti ) i = 0,..4
di Di Ti ti i = 0 1 2 3 4
45
Result
• Clear linear relation bw/ distance and time difference
• Small time offset depending on pairs.
0.85×1010cm/sec (averaged over 4 PMT pairs)
→23% lower than expected (peak speed)
46
Result
• Different data samples are analyzed.
Data sampleData sample Speed Speed [cm/s][cm/s]
Ratio to Ratio to expectationexpectation
Oct. 2008 (DRS2) 0.91×1010 0.83
Dec. 2008 (DRS2) 0.89×1010 0.81
Dec. 2008 (DRS2, high gain) 0.85×1010 0.77
Dec. 2009 (DRS4) 0.99×1010 0.90
MC 1.08×1010 0.98MC w/o scattering and
reflection1.14×1010 1.04
Vexpected ~ VMC > V2009 > V2008
47
How can the difference (2008 vs 2009)be interpreted?
• Systematics in the measurement?
• Scattering/reflection
• Fluctuation due to low photon statistics
• DRS2 vs. DRS4
• Discrepancy seems too large to explain with systematics. (systematic uncertainty ~5%)
• Speed is really changed?
• Effect of impurities
• Refractive index
• Wavelength
48
• Speed of scintillation light in LXe is measured
• Measurements in data give lower speed than expected, while MC measurement shows a good agreement with expectation.
• ~20% lower in run2008
• ~10% lower in run2009
• The difference seems too large to explain with the systematics of the measurement.
• Theoretical mechanism to change the speed is not understood yet.
The world largest xenon detector allows to measure scintillation light.The study is on going.
Analysis of 2 gamma event
50
μ → e γ γ decay
• μ → e γ γ < 7.2 x 10-11 (Crystal Box Phys.Rev.D38 2077(1988))
• In some models μ → e γ γ decay is much larger than μ → e γ
• Example : R-parity-violating (RPV) SUSY (Phys. Rev. D67 115012(2003))
• One RPV SUSY model says BR(μ → e γ γ) = 1.3 x 10-8 x |λλ’|2
• If we lower the limit, we can set the limit on RPV coupling products
μ e
scalar particle
2
51
μ → e γ γ event property
• Most probable event :
• Both photons are around 26 MeV
• Sum energy of photons is a bit higher than 52.8 MeV
• Angle between photons is small
• positron energy is a bit smaller than 52.8 MeV
• positron is around back to back to the photons
• Photons and positron are on the same plane
• Efficiency ~ 3.8 x 10-2 x Ω/4π = 3.4 x 10-3
• single event sensitivity for 2008 (Preliminary) :
• 1/Nμ/eff. = 1/9.13x1013/3.4x10-3/trig.eff./reconstruction eff. = 3.2x10-12/trig.eff./reconstruction eff.
• Sum energy of most of events < CR veto, almost back to back, photons positron at the same time → trig. eff. should not be too small
6
52
Position reconstruction of 2 photons
(Preliminary, no shower correction)
8
higher energy gamma
lower energy gamma
u, σ = 6.2 mm
u, σ = 6.2 mm
Red: higher energy gammaBlue: lower energy gamma
reconstructed energy (arbitrary)
Usual pileup rejection from time distribution can be used to reject accidental two gamma events.
53
• Estimation of trigger efficiency by MEG trigger (w/ and w/o APD)
• To use 2008 data
• Reconstruction of positron timing and angle by TC hit
• Timing reconstruction for 2 photons with better quality
• Estimation of resolution and reconstruction efficiency
• Background estimation
• 1 radiative decay + 1 photon from radiative, AIF, bremsstrahlung
• michel + 2 photons from radiative decays, AIF, bremsstrahlung
10
Development of analysis for μ → e γ γ was just started.
There are still a lot of things to check possibility of the analysis.